Proper shutdown for gcp-direct in case no documents are to process

[johann-petrak]

See https://github.com/GateNLP/gateplugin-Elasticsearch/issues/3

[ianroberts]

For reference since the linked issue is in a private repo: the problem is that if a batch involves any components - input handlers, output handlers, or PRs - that start non-daemon threads at init time, and the batch has no documents that need processing (e.g. an existing report file shows all the available document IDs as already processed successfully), then the GCP Java process will hang forever. This is because in that specific scenario the matching "close" methods on the input/output handlers and Factory.deleteResource on the application - which would typically terminate the threads started at init - are not called.

Normally the initialisation happens when building the Batch object in BatchRunner but the shutdown is triggered via PooledDocumentProcessor after parallel processing is complete. In the "no documents to process" case, no PooledDocumentProcessor is ever created.